Machine Learning In Underwater Target Recognition

The recognition of underwater vocal data is one of the most important underwater research area.The aim of this research is to find an uncontact way to recognize the target correctly based on algorithm.lt is very meaningful for both ocean development and national defense security to find a more intelligent algorithm.The machine learning algorithm contained an intelligent recognition ways that has received much attention in recent years,which provides a solution to this problem.In this paper,the ship radiation noise data in the underwater acoustic data is taken as the identification object,and the related machine learning classification and identification research is carried out.The main research contents are as follows:1?Research and analysis on the mechanism of ship radiation.Based on this,the actual measured real data is properly preprocessed,and the training and test data sets required by the subsequent algorithms are produced.In addition,the application of machine learning in high-dimensional data visualization is studied,and the distribution of experimental data is simply observed.2?The Meier cepstrum coefficient and the Shannon entropy feature extraction and fusion are performed on the target signal.The fusion feature is combined with the K-nearest neighbor classification learning algorithm and the Xgboost algorithm,and the machine learning model evaluation index is introduced to objectively evaluate the two models,which proves Xgboost can be better applied to classification and recognition of underwater acoustic data.At the same time,two kinds of machine learning dimensionality reduction algorithm principal component analysis(PCA)and neighborhood component analysis(NCA)are studied.It is proved by experiments that the NC A dimensionality reduction method can achieve a significant reduction in feature dimension while still ensuring higher performance than PCA.3?Deep learning algorithm that has received much attention in machine learning has been studied.Two deep learning methods,convolutional neural networks and long-short-time memory(LSTM)neural networks are studied.Two feature extraction methods are studied for two different neural network characteristics.Two methods of time-frequency map generation using wavelet transform and short-time Fourier transform are used for convolutional neural network research;instantaneous frequency characteristics are used for LSTM neural network research.Experiments show that the combination of wavelet transform and convolutional neural network produces the best experimental results.The combination of Xgboost algorithm and fusion feature can achieve highe recognition performance under low hardware conditions,while LSTM neural network can not achieve a good classification effect.It proves that not all deep learning algorithms are available for underwater acoustic data.